for most of it it's only funny if you're expecting it to be funny and you get thrown a curveball in an anti-joke way. the rest of the humor was when the AI fucked up, like, I was watching it and the characters just disappeared for 5 minutes and it kept cutting between scenes, which was a bit funny.
the AI itself cannot comprehend the structure and concept behind a joke. it sets up a pedestrian, everyday scenario like most sitcom humor but it either forgets its telling a joke and gets lost in playing the scenario entirely straight, or simply cannot comprehend the basic functions behind a punchline.
I think it's the latter.
I don't think machine learning is at a point where it can account for the idea of subversion, because subversion requires context, and that's as easy of a thing to teach to a model as it is to make an english language API for C++, as in, almost fucking impossible because written language is too fucking flexible and contextual to have an instructions machine truly grasp the meaning and intent behind what it spews out.
They're just more sophisticated markov chains. Data comes in, data goes out. All meaning behind what it returns is pure and utter coincidence, and that lack of true ability to cultivate understanding (vs the current system of data good data bad) is one of the fundamental barriers that I don't think the Rationalist AI-worshipper contingent in silicon valley will be able to surpass, at least for a while.
i watch expecting it to be utterly nonsensical and am rewarded when they make an actual joke. its scratches the same itch as gambling watching them get closer and closer to the punchline...
for most of it it's only funny if you're expecting it to be funny and you get thrown a curveball in an anti-joke way. the rest of the humor was when the AI fucked up, like, I was watching it and the characters just disappeared for 5 minutes and it kept cutting between scenes, which was a bit funny.
the AI itself cannot comprehend the structure and concept behind a joke. it sets up a pedestrian, everyday scenario like most sitcom humor but it either forgets its telling a joke and gets lost in playing the scenario entirely straight, or simply cannot comprehend the basic functions behind a punchline.
I think it's the latter.
I don't think machine learning is at a point where it can account for the idea of subversion, because subversion requires context, and that's as easy of a thing to teach to a model as it is to make an english language API for C++, as in, almost fucking impossible because written language is too fucking flexible and contextual to have an instructions machine truly grasp the meaning and intent behind what it spews out.
They're just more sophisticated markov chains. Data comes in, data goes out. All meaning behind what it returns is pure and utter coincidence, and that lack of true ability to cultivate understanding (vs the current system of data good data bad) is one of the fundamental barriers that I don't think the Rationalist AI-worshipper contingent in silicon valley will be able to surpass, at least for a while.
90% of the comedy in Seinfeld is delivery
i watch expecting it to be utterly nonsensical and am rewarded when they make an actual joke. its scratches the same itch as gambling watching them get closer and closer to the punchline...